CN113627544A - Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion - Google Patents
Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion Download PDFInfo
- Publication number
- CN113627544A CN113627544A CN202110935063.6A CN202110935063A CN113627544A CN 113627544 A CN113627544 A CN 113627544A CN 202110935063 A CN202110935063 A CN 202110935063A CN 113627544 A CN113627544 A CN 113627544A
- Authority
- CN
- China
- Prior art keywords
- machine tool
- source heterogeneous
- milling cutter
- data
- state
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 108
- 230000004927 fusion Effects 0.000 title claims abstract description 44
- 238000003801 milling Methods 0.000 title claims abstract description 35
- 238000004519 manufacturing process Methods 0.000 claims abstract description 35
- 238000012544 monitoring process Methods 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000000605 extraction Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000012795 verification Methods 0.000 claims abstract description 3
- 238000012549 training Methods 0.000 claims description 49
- 230000006870 function Effects 0.000 claims description 26
- 230000008569 process Effects 0.000 claims description 16
- 210000002569 neuron Anatomy 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 9
- 238000013178 mathematical model Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000010923 batch production Methods 0.000 claims description 5
- 230000006835 compression Effects 0.000 claims description 5
- 238000007906 compression Methods 0.000 claims description 5
- 238000005520 cutting process Methods 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 4
- 238000012217 deletion Methods 0.000 claims description 3
- 230000037430 deletion Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 229940050561 matrix product Drugs 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 238000012797 qualification Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000005236 sound signal Effects 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 238000010998 test method Methods 0.000 claims description 3
- 238000013145 classification model Methods 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 239000000126 substance Substances 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 7
- 238000005259 measurement Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003754 machining Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 229910001018 Cast iron Inorganic materials 0.000 description 1
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 229910045601 alloy Inorganic materials 0.000 description 1
- 239000000956 alloy Substances 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 229910002804 graphite Inorganic materials 0.000 description 1
- 239000010439 graphite Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention relates to a machine tool milling cutter state identification method based on multi-source heterogeneous data fusion, and belongs to the technical field of numerical control machine tool cutter wear monitoring. Which comprises the following steps: a data preprocessing method based on compressed sensing and noise adding processing; a stack sparse self-encoder based on Dropout method; a data fusion algorithm based on an improved D-S evidence theory; a tool state monitoring model based on SSAE and improved D-S evidence theory; the practicability of the method provided by the invention is verified, and the production line state data is acquired in an automatic monitoring platform of the production line for test verification. The method solves the problems of overfitting of a deep learning network and low identification precision of small-sample multi-source heterogeneous data. Compared with an artificial feature extraction method and an SSAE method, the method has better identification precision and generalization performance. The recognition result shows that the model and the method can accurately reflect the state of the milling cutter of the machine tool in the production line of the multi-variety small-batch aerospace enterprises.
Description
Technical Field
The invention relates to a machine tool milling cutter state identification method based on multi-source heterogeneous data fusion, and belongs to the technical field of numerical control machine tool cutter wear monitoring.
Background
The machine tool processing products of the production line of the aerospace enterprise are various in types, small in batch and complex in process and environment, so that the information and emergency situations of the machine tool processing products become complex and uncertain, and the result of monitoring the cutter by using a single-source signal is unreliable. Although the problem that single-source signal monitoring is unreliable is solved in the existing research, a deep learning network needs a large amount of training data, however, available samples are few and unbalanced in a large amount of multi-source heterogeneous data generated in the machining process of the actual machine tool cutter of a multi-variety small-batch production line, and therefore the recognition accuracy of the trained model to the state of the cutter is low.
In the process of data fusion, although the D-S (Dempster-Shafer) evidence theory performs well in a decision fusion part, the traditional D-S evidence theory cannot solve the problems that the evidence is seriously and completely conflicted, the fusion result is excessively sensitive to evidence change, and under the conditions of few samples and unbalanced categories, a certain evidence or a newly added evidence cannot influence the synthesis result of a certain event.
Disclosure of Invention
The invention provides a tool state recognition method based on multi-source heterogeneous data fusion, which is characterized in that the sparsity of samples is improved and the samples are balanced by utilizing a compressed sensing technology, the number of training samples is increased by using a noise processing method so as to improve the robustness of a network, a stack sparse self-encoder based on a Dropout (discarding method) method is provided for recognition and classification to solve the over-fitting problem of small samples, decision fusion is carried out by improving a D-S evidence theory so as to make up the defects of the traditional D-S evidence theory, tool state recognition is carried out by taking a certain aerospace enterprise production line as an example, and a test result shows that the tool wear state can be effectively recognized by the method.
Aiming at the problems in the prior art, the invention provides a tool state identification method based on multi-source heterogeneous data fusion, which is used for effectively identifying the tool wear state.
The purpose of the invention is realized by the following technical scheme.
A machine tool milling cutter state identification method based on multi-source heterogeneous data fusion comprises the following steps: s1, a data preprocessing method based on compressed sensing and noise adding processing; s2, stacking sparse self-encoder based on Dropout method; s3, a data fusion algorithm based on the improved D-S evidence theory; s4, establishing a tool state monitoring model based on SSAE (stacked sparse self-encoder) and improved D-S evidence theory; and S5, testing and verifying. The model and the algorithm provided by the invention can accurately reflect the state of the machine tool cutter of the production line of various small-batch aerospace enterprises.
According to the invention: step S1 includes the following sub-steps:
establishing a multi-source signal compression perception model of machine tool cutter wear,
and (3) carrying out noise processing on the data:
according to the invention: step S2 includes the following sub-steps:
establishing an overall cost function model of the sparse self-encoder, and calculating a formula:
establishing a softmax classifier output probability model:
according to the invention: step S5 includes the following sub-steps:
the data acquisition is carried out on the data,
the determination of the parameters is carried out,
the results of the tests were analyzed and,
the test method was compared to other methods.
Specifically, the invention provides a machine tool milling cutter state identification method based on multi-source heterogeneous data fusion, which comprises the following steps:
s1: a data preprocessing method based on compressed sensing and noise adding processing;
s2: a stack sparse self-encoder based on Dropout method;
s3: a data fusion algorithm based on an improved D-S evidence theory;
s4: a tool state monitoring model based on SSAE and improved D-S evidence theory;
s5: the practicability of the method provided by the invention is verified, and the production line state data is acquired in an automatic monitoring platform of the production line for test verification.
Further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S1 includes the following substeps:
s11: establishing a multi-source signal compression perception model of machine tool cutter wear;
s12: random Gaussian noises with different degrees are added into each group of cutter wear data training samples.
Further, in the above method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S2 includes the following substeps:
s21: calculating an overall cost function of the sparse self-encoder;
s22: on the basis of a Dropout method, the overfitting problem in the network training process is effectively reduced by randomly reducing partial connection weights of the hidden layer, so that the network is forced to learn more robust features;
s23: the method combines unsupervised learning with supervised learning to effectively extract deep features of data;
s24: and obtaining the optimal parameters through an optimization algorithm to obtain the softmax classification model.
Further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S3 includes the following substeps:
s31: determining that it satisfies the condition:
m(φ)=0
s32: establishing a mathematical model of the synthetic probability of the event A and the average support degree q (A) thereof under the same identification frame theta:
s33: establishing a combined BPA model of the event A on the basis of the step S32:
m(A)=p(A)+k×ε×q(A),A≠φ,X
define k as:
further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S4 includes the following substeps:
s41: acquiring multi-source heterogeneous data containing cutter state information, such as cutting force signals, vibration signals, sound signals and the like, from an automatic monitoring platform of a multi-variety small-batch production line of a certain aerospace enterprise;
s42: performing simple preprocessing on original multi-source heterogeneous data, including missing value processing, useless feature deletion and balancing processing;
s43: carrying out compressed sensing processing on the simply preprocessed data; then, randomly taking 80% of samples as training samples and carrying out noise adding treatment, increasing the number of the training samples, and taking the rest 20% of samples as test samples; finally normalizing all data to be in the range of [0, 1 ];
s44: initializing parameters of a deep neural network and Dropout parameters, setting a threshold value of the quality qualification rate of the lowest processed product, inputting training samples, training layer by layer, minimizing a cost function, obtaining optimal weight and bias, extracting low-dimensional features, then continuously updating the Dropout parameters, continuously training a sparse self-encoder and a softmax classifier, and finishing primary training;
s45: reversely fine-tuning the whole network by using a back propagation algorithm based on the obtained optimal weight and bias to obtain and store the optimal weight and bias;
s46: inputting test sample data into the trained network to obtain a recognition result;
s47: and (4) fusing decision layers by using an improved D-S evidence theory method based on the network identification result, and synthesizing a final identification result.
Further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S5 includes the following substeps:
s51: collecting data;
s52: obtaining and determining parameters through experiments;
s53: analyzing test results;
s54: the test methods were compared to the manual feature extraction method, SSAE, CNN and RNN.
Further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S11 includes the following substeps:
step 1: establishing a non-undersampled tool wear signal xN×1The formula is calculated as follows:
xN×1=ψN×N·SN×1
step 2: converting the sampling process into a matrix product form:
cM×1=ΦM×N·xN×1
in the formula phiM×N=[φ1,φ2,…,φN]Is an M × N observation matrix
And step 3: solving a sparse representation equation set of the machine tool cutter wear signal under the orthogonal basis matrix according to the step 1 and the step 2:
in the formula, M & lt N, the problem solved by the formula is an underdetermined problem, namely an NP-hard problem with infinite solutions, and an original signal cannot be reconstructed,
and 4, step 4: by → solving for optimal l1The norm yields the solution of the reconstructed sparse signal as:
in the formula, the former term is an error term and is used for controlling the amplitude precision of the solution result, the higher the occupation ratio is, the higher the precision is, the latter term is a regularization term and is used for controlling the sparsity of the solution result, and lambda is a regularization parameter and is a factor mainly used for balancing the error and the sparsity.
Further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, the step S12 is specifically as follows: training samples c for each set of tool wear data1(t) adding R random Gaussian noises with different degrees:
further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, step S21 includes the following substeps:
step 1: assuming that there is a multi-source heterogeneous sample set x ═ x of an unlabeled tool state1,x2,x3,…,xnThe encoder uses a coding function fθConverting a multi-source heterogeneous sample set x into a coding vector a:
a=f(W1x+b1)
step 2: using a decoding function gθThe feature vector a is transformed into y ═ y in the reverse direction1,y2,y3,…,yn}:
y=g(W2a+b2)
And step 3: applying sparsity constraints to hidden nodes:
wherein the content of the first and second substances,for the average activation of the jth hidden layer neuron under n training samples,represents a sample x(i)Degree of activation of the jth hidden layer neuron upon input;
and 4, step 4: for the training set, in order to avoid learning and improve the ability to acquire important information, the average activation amount of each hidden node j needs to be set to 0 or close to 0, so that an additional penalty term is added:
in the formula, s2The number of the neurons of the hidden layer is rho, the rho is a sparsity parameter, and the formula shows that each neuron of the hidden layer uses the sparsity parameter rho and the average activation degreeIs used to measure the difference between the two distributions when p andthe greater the difference between them, the greater its value, the greater the difference between the two distributions;
and 5: and (3) integrating the steps 1, 2, 3 and 4, wherein the overall cost function of the sparse self-encoder is as follows:
wherein, the first term is a cost loss function of the self-encoder, the second term is a sparse penalty factor term, beta is a weight coefficient of the sparse penalty factor and is used for controlling the relativity between the reconstruction term and the penalty term,
where the first term is the reconstruction error of the entire data set and the second term is the regularized weight penalty term, which aims to reduce the weight magnitude to prevent overfitting, λ represents the weight decay parameter, mlIndicating the number of network layers, sl、sl+1Represents the number of neuron nodes of the l-th layer and the l + 1-th layer,is the weight between the jth neuron at level l and the ith neuron at level l + 1.
Further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, the step S24 is specifically as follows: for a training data set formed by m training samples of k classes, the output probability of a classifier is as follows:
further, in the method for identifying the state of the milling cutter of the machine tool based on the multi-source heterogeneous data fusion, the step S31 includes the following specific steps:
step 1: mathematical model to determine evidence confidence:
step 2: on the basis of step 1, a combined BPA mathematical model of an unknown event X is established:
m(X)=p(X)+k×ε×q(X)+k×(1-ε)。
has the advantages that:
aiming at the problems of low tool utilization rate, poor production stability and the like caused by complex and uncertain process, complex and uncertain product structure, complex and uncertain information and complex and uncertain emergency of multi-variety small-batch space enterprises, the method for identifying the state of the tool of the machine tool based on fusion of multi-source heterogeneous data is provided, the multi-source data is compressed by using a compression sensing method to balance samples and improve the sparsity of the samples, and random Gaussian noise is added to the compressed data to improve the robustness of a network; a stack sparse self-encoder network based on an improved D-S evidence theory and a Dropout method is constructed, unsupervised learning is combined with supervised learning, and a Dropout method is introduced to train the network, so that the problems of overfitting of a deep learning network and the recognition accuracy of small-sample multi-source heterogeneous data are solved.
The invention provides a machine tool milling cutter state identification method based on multisource heterogeneous data fusion, which aims at solving the problems of difficult tool state identification and low precision caused by small tool wear data set samples, multisource heterogeneity and undersampling due to flexible mixed flow machining and multi-process cross parallel of various small batch production lines, and utilizes a compression sensing method to compress multisource data to balance samples and improve sample sparsity, and adds random Gaussian noise to the compressed data to improve network robustness; on the basis, a stack sparse self-encoder network based on an improved D-S evidence theory and a Dropout method is constructed, the network is trained by combining unsupervised learning with supervised learning and introducing the Dropout method, and the problems of overfitting of a deep learning network and low identification precision of small-sample multi-source heterogeneous data are solved. Finally, the state of the machine tool of the production line is identified by taking a certain aerospace enterprise production line as an example, and compared with an artificial feature extraction method and an SSAE method, the method has better identification precision and generalization performance. The recognition result shows that the model and the method can accurately reflect the state of the milling cutter of the machine tool in the production line of the multi-variety small-batch aerospace enterprises.
Drawings
FIG. 1 is a diagram of a compressed sensing linearity measurement process of a raw signal according to the present invention.
FIG. 2 is a diagram of a self-encoder of the present invention.
Fig. 3 is a diagram comparing a standard network and a network with Dropout according to the present invention.
FIG. 4 is a diagram of a 2-level stacked sparse self-encoder according to the present invention.
FIG. 5 is a block diagram of the overall process line status identification of the present invention.
FIG. 6 is a partial view of the components of the production line of the workshop, the site of the workshop and the equipment on the production line.
FIG. 7 is a diagram of an automated monitoring platform for a manufacturing shop according to the present invention.
FIGS. 8(a) -8 (c) are graphs of signals of the power, cutting force and vibration signals of the new cutter of the present invention, respectively.
FIG. 9 is a graph of the error of the present invention at different measurement lengths.
FIGS. 10(a) -10(f) are graphs illustrating the effect of various parameters of the network of the present invention on the recognition accuracy.
FIGS. 11(a) and 11(b) are graphs of the recognition accuracy and loss function values of the training and testing of the deep neural network of the present invention.
FIGS. 12(a) -12(c) are scatter plots of the extracted features of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following drawings and examples, but it should be understood that the examples are illustrative of the present invention and are not intended to limit the present invention.
The non-undersampled tool wear signal can be expressed as:
xN×1=ψN×N·sN×1
converting the sampling process into a matrix product form:
cM×1=ΦM×N·xN×1
solving a sparse representation equation set of the machine tool cutter wear signal under the orthogonal basis matrix:
at a certain orthogonal basis matrix psiN×NUpper xN×1Is sparse, if phiM×NSatisfy constraint equidistance condition, can solve optimum l by1The norm yields the solution of the reconstructed sparse signal as:
as shown in fig. 1, in a compressed sensing linear measurement process of an original signal, after a sparse representation of a production line under-sampled state signal is obtained, the sparse representation is substituted into an equation, and then reconstruction of the production line non-under-sampled state signal can be achieved.
In order to improve the robustness of the network and reduce the interference sensitivity of the network to the tool wear data, a training sample c needs to be trained for each group of tool wear data in advance1(t) adding R random Gaussian noises with different degrees:
after noise is added, the number of samples of the tool wear data of each group of machine tools is expanded into R +1 groups of samples
The basic network structure of the self-encoder is shown in fig. 2. The self-encoder is a network which reconstructs input data through a network and realizes unsupervised feature extraction and parameter learning. The conversion process from the input layer to the hidden layer is an encoding process that functions to extract characteristic information from the input data. The conversion process from the hidden layer to the output layer is a decoding process that functions to reconstruct the feature information into input data.
Assuming that there is a multi-source heterogeneous sample set x ═ x of an unlabeled tool state1,x2,x3,…,xnThe encoder uses a coding function fθAnd converting the multi-source heterogeneous sample set x into a coding vector a.
a=f(W1x+b1)
The decoder uses a decoding function gθThe feature vector a is transformed into y ═ y in the reverse direction1,y2,y3,…,yn}。
y=g(W2a+b2)
Applying sparsity constraint to the hidden nodes, and calculating a formula:
for the training set, in order to avoid learning and improve the ability to acquire important information, it is necessary to set the average activation amount of each hidden node j to 0 or close to 0. Therefore, an additional penalty is added:
the overall cost function of the sparse autoencoder is:
as shown in fig. 3, a part of hidden layer neurons are forced to stop working randomly, and the structure of the whole network is already different, the Dropout method can effectively reduce the over-fitting problem in the network training process by randomly reducing the part of connection weights of the hidden layer, so as to force the network to learn more robust features.
The stack sparse autoencoder is to remove the output layer of the sparse autoencoder, only reserve the coding part, use the feature data extracted from the coding part as the input of the next sparse autoencoder, and connect a plurality of sparse autoencoders by the same method, thereby extracting the feature information with higher dimension, fig. 4 is a structure diagram of the 2-layer stack sparse autoencoder.
Let the training set be { (x)(1),y(1)),(x(2),y(2)),…,(x(n),y(n))},x(i)Is sample data, y(i)For the sample label, the output probability of the classifier is:
based on the improved D-S evidence theory which is improved by improving Yager. Let the recognition frame theta of the event A and the corresponding evidence body be F, if the function m, 2Θ→[0,1]It satisfies the condition:
m(φ)=0
the synthetic probability of the event a and the average support degree q (a) thereof under the same recognition framework Θ are:
on this basis, the combination of event A BPA is
m(A)=p(A)+k×ε×q(A),A≠φ,X
k is a normalization constant, i.e. a collision factor, representing the degree of collision between evidences, defined as
The evidence confidence, ε, the confidence level of the evidence, may be expressed as:
n is the number of evidences, kijRepresentative evidence FiAnd evidence FjThe conflict between them can be expressed as:
considering that the conflict cannot be determined, introducing an unknown event X, and assigning the conflict to the unknown event X, wherein the combination BPA of the unknown event X is as follows:
m(X)=p(X)+k×ε×q(X)+k×(1-ε)
the invention gives full play to the advantages of deep learning, combines the compressed sensing technology and the improved D-S evidence theory, and provides an effective method for monitoring the state of the machine tool cutter of the production line of the multi-variety small-batch aerospace enterprises, the overall framework of the method is shown in figure 5, and the specific algorithm steps are as follows:
step 1: the method comprises the steps of obtaining multi-source heterogeneous data containing cutter state information, such as cutting force signals, vibration signals, sound signals and the like, from an automatic monitoring platform of a multi-variety small-batch production line of a certain aerospace enterprise.
Step 2: and carrying out simple preprocessing on the original multi-source heterogeneous data, including missing value processing, useless feature deletion and balancing processing.
And step 3: carrying out compressed sensing processing on the simply preprocessed data; then, randomly taking 80% of samples as training samples and carrying out noise adding treatment, increasing the number of the training samples, and taking the rest 20% of samples as test samples; finally all data were normalized to the [0, 1] range.
And 4, step 4: initializing parameters of a deep neural network and Dropout parameters, setting a threshold value of the quality qualified rate of the lowest processed product, inputting training samples, training layer by layer, minimizing a cost function, obtaining optimal weight and bias, and extracting low-dimensional features. And then continuously updating Dropout parameters, and continuously training the sparse self-encoder and the softmax classifier to finish primary training.
And 5: and reversely fine-tuning the whole network by utilizing a back propagation algorithm based on the obtained optimal weight and bias to obtain and store the optimal weight and bias.
And inputting the test sample data into the trained network to obtain a recognition result.
And (4) fusing decision layers by using an improved D-S evidence theory method based on the network identification result, and synthesizing a final identification result.
As shown in fig. 6, a certain aerospace enterprise mainly produces 8 kinds of complex aerospace components such as a fixed frame, a shell, a fuse holder, a servo mechanism rotating frame, a channel body, an oil cylinder body, an antenna main cavity body and a machine base, and has the disadvantages of complex precision of parts, complex processing technology, various types of related processing equipment, complex operation and complex production conditions. The production line consists of 76 devices such as a vertical machining center, a numerical control vertical milling machining center, a drilling and milling machining center, a numerical control turning center, a numerical control lathe, a precise numerical control lathe, a horizontal numerical control lathe, a coordinate boring and milling machining center and the like
As shown in fig. 7, the production line of the multi-variety small-lot space enterprises adopts an automatic monitoring platform of a certain domestic company, and the automatic monitoring platform provides a very comprehensive workshop monitoring function, can effectively monitor the states of the production line and equipment, and effectively visualizes the current state of the production line with information. The monitoring platform has the informationized monitoring of a workshop level and a unit level, can control any unit equipment on a production line by automatically and manually switching, and has good management on the cutter.
As shown in fig. 8(a) -8 (c), the present invention obtains data of a creep graphite cast iron part processed by using a ball head hard alloy milling cutter (three cutters) in a german maho 1600S double-processing vertical and horizontal processing center during processing, and when the obtained data is a constant milling process, multi-source heterogeneous data such as a machine tool spindle power signal, a product processing qualification rate, a cutting force signal, a vibration signal, an acoustic emission signal and the like are collected to obtain 400 samples in total.
As shown in fig. 9, the measurement length M is 500 from the reconstructed error trend graph. After all samples are subjected to compressed sensing resampling, the data point number of each sample is compressed to 500 points, 50% of samples are randomly taken as training data, 3 random Gaussian noises are added to the training data, the noise intensity coefficient k is 0.1, and the number of the training samples is expanded by 3 times.
SSAE increases the depth of the network by increasing the number of hidden layers, so as to extract deeper features, and SSAEs with different numbers of layers have different accuracy in signal identification under the condition that other conditions are unchanged (as shown in fig. 10 (a)). Similarly, the experimental results of the learning rate, the sparse parameter, the weight penalty parameter, the sparse weight penalty parameter, and the Dropout parameter under the condition that other conditions are kept unchanged are shown in fig. 10(b), 10(c), 10(d), 10(e), and 10 (f).
Other network parameter settings are shown in Table 1
Table 1 other network parameter settings
And after the network hyper-parameter is set, setting a program to stop running in advance when the reconstruction error is stable, inputting the training set into the deep neural network for training, saving the optimal weight and bias after the training is finished, and inputting the test set to verify the performance of the sparse self-encoder. The experimental results obtained by the trial and error are shown in fig. 11(a) and 11 (b).
Setting a production line state space as an identification frame: Θ ═ initial wear of tool and normal machine tool a1Medium wear of the tool and normal operation of the machine tool A2Severe wear of the tool and normal operation of the machine tool A3Tool breakage and machine tool normality A4Machine tool anomaly A5And the evidence bodies are recognition results of various data, a basic probability distribution function of each evidence body is solved, the fused trust degree is obtained according to an improved D-S evidence theory fusion rule, final decision is finally made according to a rule of a maximum trust function, and the final decision result is shown in a table 2.
TABLE 2 production line state recognition method decision layer fusion results based on D-S evidence theory
The recognition accuracy of 800 training samples in 5 states after decision layers of improved D-S evidence theory are fused reaches 94.87%, the recognition accuracy of 200 test samples also reaches 99%, the effectiveness of a deep learning recognition model based on the improved D-S evidence theory in the aspect of fusion of multi-source data to production line state decision is fully verified, the production line state information in different forms contained in multi-source signals is effectively utilized, the advantages of the signals are fully exerted, mutual complementation is realized, the effective fusion of the multi-source signals is realized, and the expected target of production line state monitoring is achieved.
To verify the effectiveness of the method proposed by the present invention, it was compared with the manual feature extraction method, SSAE, CNN and RNN. The artificial feature extraction method extracts 10 features in total, including a maximum value, a root mean square value, a skewness value, a kurtosis value, a waveform index, a pulse index, a skewness index, a peak index, a margin index and a kurtosis index. Inputting 10 characteristics into an SVM, setting a penalty factor to be 3, adopting an RBF (radial basis function) as a kernel function, and setting kernel function parameters to be 1; the CNN and RNN method sets the learning rate to be 0.1, the iteration number to be 400, the training batch to be 5 and the network structure to be [2048, 512, 256, 5 ]; the input of the SSAE is data which is not compressed by a CS method, the network is set to [512, 256, 128, 5], and other hyper-parameters are consistent with those set in the SSAE model of the invention; the CS + SSAE method set-up is consistent with the test set-up of the present invention. In order to reduce accidental errors in the tests, 10 tests were carried out in each method, and the average value of the results of the 10 tests was taken as an index for comparison. The recognition effect of each method is shown in table 3.
TABLE 3 training test recognition accuracy and training time for different algorithms
In order to verify the feature extraction capability of the proposed method, the first three principal component components of the proposed method, the artificial feature extraction method and the SSAE method of the present invention were extracted and visualized using Principal Component Analysis (PCA), respectively, as shown in fig. 12(a) -12 (b). As can be seen from the figure, it is difficult to distinguish the features of the 5 states using the manual feature extraction method and the SSAE method, and the extracted features are crossed and cannot be identified. The method provided by the invention can well gather the characteristics of the same state together, and different states are effectively separated, so that the characteristic extraction capability of the method provided by the invention is fully embodied.
In conclusion, compared with the identification method directly using the stack sparse autoencoder, the method for identifying the production line state of the space enterprise based on the multi-source heterogeneous data fusion has higher identification precision. Compared with an artificial feature extraction method and a stack sparse self-encoder, the method provided by the invention has stronger feature extraction capability. Compared with the traditional production line state identification method, the method provided by the invention is more intelligent, can be used for extracting effective information contained in the production line state signal in a self-adaptive manner, gets rid of dependence on a large amount of data processing knowledge and priori knowledge, improves the precision and reliability of production line state identification, and has better generalization capability and reliability.
The above description is a preferred embodiment of the present invention, and is not intended to limit the present invention, and those skilled in the art may modify the above technical solutions or substitute some technical features of the above technical solutions. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (11)
1. A machine tool milling cutter state identification method based on multi-source heterogeneous data fusion is characterized by comprising the following steps:
s1: a data preprocessing method based on compressed sensing and noise adding processing;
s2: a stack sparse self-encoder based on Dropout method;
s3: a data fusion algorithm based on an improved D-S evidence theory;
s4: a tool state monitoring model based on SSAE and improved D-S evidence theory;
s5: the practicability of the method provided by the invention is verified, and the production line state data is acquired in an automatic monitoring platform of the production line for test verification.
2. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion according to claim 1, wherein the step S1 comprises the following sub-steps:
s11: establishing a multi-source signal compression perception model of machine tool cutter wear;
s12: random Gaussian noises with different degrees are added into each group of cutter wear data training samples.
3. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion according to claim 1, wherein the step S2 comprises the following sub-steps:
s21: calculating an overall cost function of the sparse self-encoder;
s22: on the basis of a Dropout method, the overfitting problem in the network training process is effectively reduced by randomly reducing partial connection weights of the hidden layer, so that the network is forced to learn more robust features;
s23: the method combines unsupervised learning with supervised learning to effectively extract deep features of data;
s24: and obtaining the optimal parameters through an optimization algorithm to obtain the softmax classification model.
4. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion as claimed in claim 2, wherein the step S3 comprises the following sub-steps:
s31: determining that it satisfies the condition:
m(φ)=0
s32: establishing a mathematical model of the synthetic probability of the event A and the average support degree q (A) thereof under the same identification frame theta:
s33: establishing a combined BPA model of the event A on the basis of the step S32:
m(A)=p(A)+k×ε×q(A),A≠φ,X
define k as:
5. the method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion according to claim 1, wherein the step S4 comprises the following sub-steps:
s41: acquiring multi-source heterogeneous data containing cutter state information, such as cutting force signals, vibration signals, sound signals and the like, from an automatic monitoring platform of a multi-variety small-batch production line of a certain aerospace enterprise;
s42: performing simple preprocessing on original multi-source heterogeneous data, including missing value processing, useless feature deletion and balancing processing;
s43: carrying out compressed sensing processing on the simply preprocessed data; then, randomly taking 80% of samples as training samples and carrying out noise adding treatment, increasing the number of the training samples, and taking the rest 20% of samples as test samples; finally normalizing all data to be in the range of [0, 1 ];
s44: initializing parameters of a deep neural network and Dropout parameters, setting a threshold value of the quality qualification rate of the lowest processed product, inputting training samples, training layer by layer, minimizing a cost function, obtaining optimal weight and bias, extracting low-dimensional features, then continuously updating the Dropout parameters, continuously training a sparse self-encoder and a softmax classifier, and finishing primary training;
s45: reversely fine-tuning the whole network by using a back propagation algorithm based on the obtained optimal weight and bias to obtain and store the optimal weight and bias;
s46: inputting test sample data into the trained network to obtain a recognition result;
s47: and (4) fusing decision layers by using an improved D-S evidence theory method based on the network identification result, and synthesizing a final identification result.
6. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion according to claim 1, wherein the step S5 comprises the following sub-steps:
s51: collecting data;
s52: obtaining and determining parameters through experiments;
s53: analyzing test results;
s54: the test methods were compared to the manual feature extraction method, SSAE, CNN and RNN.
7. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion as claimed in claim 2, wherein the step S11 comprises the following sub-steps:
step 1: establishing a non-undersampled tool wear signal xN×1The formula is calculated as follows:
xN×1=ψN×N·SN×1
step 2: converting the sampling process into a matrix product form:
cM×1=ΦM×N·xN×1
in the formula phiM×N=[φ1,φ2,…,φN]Is an M × N observation matrix
And step 3: solving a sparse representation equation set of the machine tool cutter wear signal under the orthogonal basis matrix according to the step 1 and the step 2:
in the formula, M & lt N, the problem solved by the formula is an underdetermined problem, namely an NP-hard problem with infinite solutions, and an original signal cannot be reconstructed,
and 4, step 4: by → solving for optimal l1The norm yields the solution of the reconstructed sparse signal as:
in the formula, the former term is an error term and is used for controlling the amplitude precision of the solution result, the higher the occupation ratio is, the higher the precision is, the latter term is a regularization term and is used for controlling the sparsity of the solution result, and lambda is a regularization parameter and is a factor mainly used for balancing the error and the sparsity.
8. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion as claimed in claim 2, wherein the specific method of the step S12 is as follows: training samples c for each set of tool wear data1(t) adding R random Gaussian noises with different degrees:
9. the method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion according to claim 3, wherein the step S21 comprises the following sub-steps:
step 1: assuming that there is a multi-source heterogeneous sample set x ═ x of an unlabeled tool state1,x2,x3,…,xnThe encoder uses a coding function fθConverting a multi-source heterogeneous sample set x into a coding vector a:
a=f(W1x+b1)
step 2: using a decoding function gθThe feature vector a is transformed into y ═ y in the reverse direction1,y2,y3,…,yn}:
y=g(W2a+b2)
And step 3: applying sparsity constraints to hidden nodes:
wherein the content of the first and second substances,for the average activation of the jth hidden layer neuron under n training samples,represents a sample x(i)Degree of activation of the jth hidden layer neuron upon input;
and 4, step 4: for the training set, in order to avoid learning and improve the ability to acquire important information, the average activation amount of each hidden node j needs to be set to 0 or close to 0, so that an additional penalty term is added:
in the formula, s2The number of the neurons of the hidden layer is rho, the rho is a sparsity parameter, and the formula shows that each neuron of the hidden layer uses the sparsity parameter rho and the average activation degreeIs used to measure the difference between the two distributions when p andthe greater the difference between them, the greater its value, the greater the difference between the two distributions;
and 5: and (3) integrating the steps 1, 2, 3 and 4, wherein the overall cost function of the sparse self-encoder is as follows:
wherein, the first term is a cost loss function of the self-encoder, the second term is a sparse penalty factor term, beta is a weight coefficient of the sparse penalty factor and is used for controlling the relativity between the reconstruction term and the penalty term,
where the first term is the reconstruction error of the entire data set and the second term is the regularized weight penalty term, which aims to reduce the weight magnitude to prevent overfitting, λ represents the weight decay parameter, mlIndicating the number of network layers, sl、sl+1The number of neuron nodes representing the l-th layer and the l + 1-th layerThe number of the first and second groups is,is the weight between the jth neuron at level l and the ith neuron at level l + 1.
10. The method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion as claimed in claim 3, wherein the specific method of the step S24 is as follows: for a training data set formed by m training samples of k classes, the output probability of a classifier is as follows:
11. the method for identifying the milling cutter state of the machine tool based on the multi-source heterogeneous data fusion as claimed in claim 4, wherein the specific steps of the step S31 are as follows:
step 1: mathematical model to determine evidence confidence:
step 2: on the basis of step 1, a combined BPA mathematical model of an unknown event X is established:
m(X)=p(X)+k×ε×q(X)+k×(1-ε)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110935063.6A CN113627544B (en) | 2021-08-16 | 2021-08-16 | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110935063.6A CN113627544B (en) | 2021-08-16 | 2021-08-16 | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113627544A true CN113627544A (en) | 2021-11-09 |
CN113627544B CN113627544B (en) | 2024-03-08 |
Family
ID=78385498
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110935063.6A Active CN113627544B (en) | 2021-08-16 | 2021-08-16 | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113627544B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116204850A (en) * | 2023-03-14 | 2023-06-02 | 匀熵智能科技(无锡)有限公司 | Multi-mode emotion analysis method based on dynamic gradient and multi-view collaborative attention |
CN117226600A (en) * | 2023-11-14 | 2023-12-15 | 四川工程职业技术学院 | Cutter abrasion prediction method, device, medium and equipment in alloy milling |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106559744A (en) * | 2016-11-24 | 2017-04-05 | 湖北文理学院 | It is a kind of based on sparse identification and the body area network power-economizing method of compressed sensing |
CN107016241A (en) * | 2017-04-05 | 2017-08-04 | 重庆交通大学 | Based on the rotating machinery lifetime stage recognition methods for adding sample enhancing depth own coding learning network of making an uproar |
CN108764064A (en) * | 2018-05-07 | 2018-11-06 | 西北工业大学 | SAR Target Recognition Algorithms based on Steerable filter device and self-encoding encoder |
CN110070118A (en) * | 2019-04-10 | 2019-07-30 | 广东电网有限责任公司 | A kind of multi-space data fusion method |
CN112329855A (en) * | 2020-11-05 | 2021-02-05 | 华侨大学 | Underdetermined working modal parameter identification method and detection method based on adaptive dictionary |
US20210197335A1 (en) * | 2019-12-26 | 2021-07-01 | Dalian University Of Technology | Data Augmentation Method Based On Generative Adversarial Networks In Tool Condition Monitoring |
-
2021
- 2021-08-16 CN CN202110935063.6A patent/CN113627544B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106559744A (en) * | 2016-11-24 | 2017-04-05 | 湖北文理学院 | It is a kind of based on sparse identification and the body area network power-economizing method of compressed sensing |
CN107016241A (en) * | 2017-04-05 | 2017-08-04 | 重庆交通大学 | Based on the rotating machinery lifetime stage recognition methods for adding sample enhancing depth own coding learning network of making an uproar |
CN108764064A (en) * | 2018-05-07 | 2018-11-06 | 西北工业大学 | SAR Target Recognition Algorithms based on Steerable filter device and self-encoding encoder |
CN110070118A (en) * | 2019-04-10 | 2019-07-30 | 广东电网有限责任公司 | A kind of multi-space data fusion method |
US20210197335A1 (en) * | 2019-12-26 | 2021-07-01 | Dalian University Of Technology | Data Augmentation Method Based On Generative Adversarial Networks In Tool Condition Monitoring |
CN112329855A (en) * | 2020-11-05 | 2021-02-05 | 华侨大学 | Underdetermined working modal parameter identification method and detection method based on adaptive dictionary |
Non-Patent Citations (2)
Title |
---|
DONGLAN LIU 等: "Multi-source Log Comprehensive Feature Extraction Method Based on Restricted Boltzmann Machine in Power Information System", 《2019 IEEE 11TH INTERNATIONAL CONFERENCE ON COMMUNICATION SOFTWARE AND NETWORKS》, 31 December 2019 (2019-12-31), pages 503 - 508 * |
冀振燕 等: "基于深度学习的融合多源异构数据的推荐模型", 《北京邮电大学学报》, vol. 42, no. 6, 31 December 2019 (2019-12-31), pages 35 - 42 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116204850A (en) * | 2023-03-14 | 2023-06-02 | 匀熵智能科技(无锡)有限公司 | Multi-mode emotion analysis method based on dynamic gradient and multi-view collaborative attention |
CN116204850B (en) * | 2023-03-14 | 2023-11-03 | 匀熵智能科技(无锡)有限公司 | Multi-mode emotion analysis method based on dynamic gradient and multi-view collaborative attention |
CN117226600A (en) * | 2023-11-14 | 2023-12-15 | 四川工程职业技术学院 | Cutter abrasion prediction method, device, medium and equipment in alloy milling |
Also Published As
Publication number | Publication date |
---|---|
CN113627544B (en) | 2024-03-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11761930B2 (en) | Prediction method of part surface roughness and tool wear based on multi-task learning | |
CN110561191B (en) | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder | |
CN111633467B (en) | Cutter wear state monitoring method based on one-dimensional depth convolution automatic encoder | |
CN113255848B (en) | Water turbine cavitation sound signal identification method based on big data learning | |
CN111832647A (en) | Abnormal flow detection system and method | |
CN110070060B (en) | Fault diagnosis method for bearing equipment | |
CN113664612A (en) | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network | |
CN108581633B (en) | Method for optimizing multi-sensor monitoring cutter state based on genetic algorithm | |
CN113627544A (en) | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion | |
CN111539152B (en) | Rolling bearing fault self-learning method based on two-stage twin convolutional neural network | |
CN111761409A (en) | Multi-sensor numerical control machine tool cutter wear monitoring method based on deep learning | |
CN112529341B (en) | Drilling well leakage probability prediction method based on naive Bayesian algorithm | |
CN114619292B (en) | Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network | |
CN111126255A (en) | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm | |
CN110297469B (en) | Production line fault judgment method based on resampling integrated feature selection algorithm | |
CN113188794B (en) | Gearbox fault diagnosis method and device based on improved PSO-BP neural network | |
CN113798920A (en) | Cutter wear state monitoring method based on variational automatic encoder and extreme learning machine | |
CN111175046A (en) | Rolling bearing fault diagnosis method based on manifold learning and s-k-means clustering | |
CN114509266A (en) | Bearing health monitoring method based on fault feature fusion | |
CN113052271A (en) | Biological fermentation data prediction method based on deep neural network | |
Oberle et al. | A use case to implement machine learning for life time prediction of manufacturing tools | |
CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
CN112801139B (en) | Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning | |
CN113485244A (en) | Numerical control machine tool control system and method based on cutter wear prediction | |
CN108414228B (en) | Based on averagely more granularity decision rough sets and NNBC Method for Bearing Fault Diagnosis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |